Max Gekk created SPARK-59545:
--------------------------------
Summary: AvroSchemaHelperSuite map-key overflow test fails
deterministically on Java 25
Key: SPARK-59545
URL: https://issues.apache.org/jira/browse/SPARK-59545
Project: Spark
Issue Type: Bug
Components: SQL
Affects Versions: 4.3.0
Reporter: Max Gekk
h3. Problem
Since the follow-up to SPARK-59311 (apache/spark#58722, merged 2026-09-11), one
test in {{AvroSchemaHelperSuite}} fails on every Java 25 run of
{{build_java25.yml}} on master (2026-09-13 and 2026-09-15), while the default
Java 17 job passes:
{code}
- SPARK-59311: a pathologically nested map-key type names the map-key property
on overflow *** FAILED ***
java.lang.IndexOutOfBoundsException: Index -1 out of bounds for length 0 was
not instance of
org.apache.spark.sql.avro.IncompatibleSchemaException expected
IncompatibleSchemaException
{code}
The test parses a Catalyst type nested 1000 deep from the
{{spark.sql.catalyst.mapKey.type}} property on a 256 KB thread stack and
expects the StackOverflowError to surface as {{IncompatibleSchemaException}}
naming that property. On JDK 25 the overflow is absorbed somewhere inside the
ANTLR parse and what escapes is an {{IndexOutOfBoundsException}} from the
parser's own state, so the conversion in {{SchemaConverters}} never sees a
{{StackOverflowError}}.
The sibling test "a pathologically nested Catalyst type fails as a schema
error, not StackOverflowError" passes on the same JDK, so the difference is the
map-key path.
h3. Reproduction
JDK 25 (openjdk 25.0.4), x86_64, three of three runs:
{code}
build/sbt "avro/testOnly *AvroSchemaHelperSuite"
Tests: succeeded 9, failed 1
{code}
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
